SANDY BRIDGE WAS shaping up to be the killer CPU of the year, a huge step forward in the ‘uncore’, decent graphics and big gains in the core as well. Instead, we got broken graphics, non-working feature sets, and a showstopper bug. What a shattering disappointment.
By the time this goes up, there will be a ton of glowing reviews out there that never step outside of the mainstream testing, that means Windows 7 and some very carefully controlled tests. If you do, things flat out don’t work, Sandy Bridge is simply not ready for release. I don’t blame Intel for handing out favorable testing materials, everyone does that, I do however blame the review sites for toeing the line, it ill serves the community.
That said, what am I talking about? If you try to use Sandy Bridge under Linux, it is simply broken. We tried to test an Intel DH67BL (Bearup Lake) with 2GB of Kingston HyperX DDR3, an Intel 32GB SLC SSD, and a ThermalTake Toughpower 550W PSU. At first we tried to install vanilla Ubuntu 10.10/AMD64 from a Kingston Datatraveler Ultimate 32GB USB3 stick. The idea was that it would speed things up significantly on install.
That’s when the crippling bug surfaced. It seems the USB3 ports on the Intel DH67BL don’t want to work. Ubuntu 10.10 installs fail during the install, no fix was found. Plug the same stick into a USB2 port, and it works fine. Alternately, install from a USB2 stick on a USB3 port, and things work fine.
The USB3 port uses a bog standard NEC chip, so there should not be any problems, but something is badly off. The first thought is that this is a driver issue, but since it is during an OS install, the port should be seen as a USB2 port and just work at the lower speeds. It doesn’t fall back correctly, the install just blows up. Moving to either a USB2 stick or a USB2 port fixes the problem.
From there, the next biggest thing is graphics. The press slide deck from Intel is 95 pages long, 78 if you don’t count the title pages, disclaimers or fine print. Of those, 37 (47%) are all about, or prominently mention graphics or advanced graphics features. Intel thinks very highly of Sandy Bridge’s graphics.
Unfortunately for Intel, they simply do not work. No, you are not reading that wrong, the graphics are simply broken on Linux. While this is not the only thing broken, being able to put up anything more than basic 2D VGA images is rather mandatory now, and Intel flat out fails. There are no drivers available for Sandy Bridge upon release. If you buy a Sandy setup and want to run Linux, you just bought a doorstop.
One uses Sandy Bridge GPUs, the other discrete. Can you spot the difference?
It is honestly that bad. Running the Phoronix Test Suite’s OpenGL tests, the few that didn’t outright fail ran at zero to one FPS. If you plugged in a discrete card, in our case an ATI HD5830, the same tests ran at hundreds of FPS, showing that they are not exactly demanding. 600FPS on a high end card should result in fairly playable frame rates on even the wimpiest of integrated GPUs.
OpenGL on Sandy Bridge is totally non-functional. How bad is it? Well, if you turn Compiz on, moving windows around the screen is an exercise in watching things stutter. It is painful. Games as complex as Tux Racer produce a lower frame rate than the same thing on an iPhone, so slow that it is unplayable. If moving a window on the desktop or playing Tux Racer is too taxing….. Ouch. Fail.
Moving on the the video encode/decode, they were promised for the Q1 release of Intel’s open graphics drivers, found here. The jury is out on those, but for release, it is not available. That is the up side of the feature set.
The down side is pretty painful. Intel offers a bunch of must have, make your life better, brings world peace, flowers, and the like level features on Sandy Bridge. The working count on Linux? Zero on launch. The chance that they will ever work on Linux? Just slightly above zero, think the need for extended floating point precision to see that many zeros.
Let me make one technical point, most of these features are considered chipset features according to Intel, and that may be the case. That said, there is one company making Sandy Bridge chipsets, that would be Intel. The entire run of chipsets are based off the same basic architecture. Since there is no other choice, Sandy’s feature set includes the Intel 6x chipset feature set, they are inseparable.
So, back to the ‘must have or your coffee shop experience is not complete’ feature set. They include Video Processing Accelerators – never coming to Linux, Color Processing Accellerators – never coming to Linux, Skin Tone Enhancements – never coming to Linux, Adaptive Contrast Enhancment – never coming to Linux, Total Color Control – never coming to Linux, Video Decode in hardware – Q1, Video Encode in hardware – Q1, 3D accelleration – Q1 sooner rather than later and a host of software to use it – never coming to Linux.
To say that this looks grim is quite the understatement. Literally all of the graphics features barring putting a 2D image on the screen don’t work, but some may soon. Most of the rest will never come, and that is sad. Until then, Sandy Bridge is a massive regression from the previous Clarksdale and Arrandale Intel CPU/GPU chips, it simply doesn’t work.
Until Intel puts out working drivers that actually use the features of the chip, you will be buying a lot of silicon that you can not use. Given the Intel track record on graphics drivers, wait until they are released until you spend your money, their success rate is summarily awful, but at least there is some hope for basic functionality.
Normally I would recommend buying an AMD chip, they have working drivers for both the current 8xx chipsets and the upcoming Fusion parts, but lately their forward momentum seems to have come to a screeching halt. Video encode/decode is not even on the roadmaps after years in hardware, but the rest has been fairly solid, occasionally going as far as improving. HD6000 DRM (Direct Rendering Manager – not the evil stuff) is still absent, but that is imminent. With AMD, so far, so barely adequate to mediocre.
The saddest part about this is that if you plug a discrete card in, the CPU core itself seems really damn fast, unfortunately, the broken graphics are enough to overwhelm those good points. AMD has a mediocre core (K10.x) with solid graphics. Until now Intel has had a very good core with barely adequate graphics. Making the core faster however does not make up for non-functional graphics, and that is the shame of it all.
Back to the advice part, the only thing I can say is don’t go there. There is no really worthwhile alternative either, neither AMD’s 8xx chipsets nor the Intel Arrandale/Clarksdale twins can be called well supported, but of the two, Intel’s solution is head and shoulders above AMDs. If you have an older chip, probably the best thing you can do right now is nothing, don’t upgrade. Until Sandy Bridge gets fixed to at least work, avoid it.S|A
Editor’s note: Intel provided S|A with the following response.
Intel: ‘Charlie tested unreleased Sandybridge hardware without drivers. No hardware can function properly without drivers on any operating system, and as such, these results are not valid. We encourage Semiaccurate to retest the platform with the appropriate drivers in place’
Editor would like to note that Intel provided the unreleased hardware and drivers for Windows 7, Windows XP, and Vista. No drivers were provided for any flavor of Linux, and none were available short of building from source on our own. We do not feel this meets any reasonable standard for ‘available’. We await appropriate drivers from Intel for re-testing, but as of press time, none were available.
Latest posts by Charlie Demerjian (see all)
- ARM upgrades realtime offerings to v8-R and adds Cortex-R52 - Sep 21, 2016
- Everspin and Globalfoundries team up for embedded ST-MRAM - Sep 15, 2016
- Intel’s Xpoint is pretty much broken - Sep 12, 2016
- ARM adds 2048-bit vectors to v8A with SVE - Sep 7, 2016
- AMD releases Bristol Ridge 7th Generation APU - Sep 5, 2016